计算机科学与探索 ›› 2021, Vol. 15 ›› Issue (7): 1310-1321.DOI: 10.3778/j.issn.1673-9418.2005037

• 图形图像 • 上一篇    下一篇

运动信息优化相关滤波的多目标跟踪算法

缪佳妮,杨金龙,程小雪,葛洪伟   

  1. 江南大学 人工智能与计算机学院,江苏 无锡 214122
  • 出版日期:2021-07-01 发布日期:2021-07-09

Multi-target Tracking Algorithm Based on Motion Information Optimized Correl-ation Filtering

MIAO Jiani, YANG Jinlong, CHENG Xiaoxue, GE Hongwei   

  1. School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2021-07-01 Published:2021-07-09

摘要:

在结合检测器检测信息的多目标跟踪任务中,目标漏检通常会导致目标漏跟,增加目标身份标签变换等问题,从而降低跟踪精度。针对该问题,提出了一种运动信息优化相关滤波的多目标跟踪算法。该算法在得到目标的检测信息后,采用核相关滤波(KCF)对目标进行跟踪,并融入目标的运动信息和图像信息,以处理检测器结果不精确,出现大量漏跟失跟问题,减少碎片化的轨迹。同时在核相关滤波的基础上引入置信图的平滑约束来评估目标被遮挡程度,实现核相关滤波中目标模板的自适应更新,处理目标由于遮挡而产生模板污染问题。最终在MOT Challenge的MOT17数据集上的实验结果表明,与传统的检测跟踪算法IOU17相比,在多目标跟踪正确度(MOTA)指标上提高了2.43%,具有更好的稳定性和精确度。

关键词: 相关滤波, 检测跟踪, 平滑滤波, 模板更新

Abstract:

In multi-target tracking tasks combined with detector detection information, missing detections often lead to some targets missed, target identity tag conversion, etc., thereby reducing tracking accuracy. To solve this problem, a multi-target tracking algorithm based on motion information optimization and correlation filter is proposed. After obtaining target detection information, kernelized correlation filter (KCF) is used to track target, and the target??s motion information and image information are integrated to handle the problem of missing tracking due to inaccu-rate detection, reducing the fragmented trajectory. At the same time, the smoothing constraint of confidence map is introduced on the basis of KCF to evaluate occlusion degree of targets, which achieves the adaptive update of target template in KCF and deals with the problem of template pollution caused by occlusion. Finally, the experimental results on the MOT Challenge MOT17 data set show that compared with the traditional detection and tracking algor-ithm, high-speed tracking-by-detection without using image information (IOU17), multiple object tracking accuracy (MOTA) of proposed algorithm is improved by 2.43%, and it has better stability and accuracy.

Key words: correlation filter, detection and tracking, smoothing filtering, template update